Manual — How to Use Every Page

SECTION 3 — PROCEDURES

Step by step, in the order you actually do them. Every underlined term links to the page it happens on.

3.1 — SET UP YOUR ACCOUNT

Goal: an account that can actually generate.

Sign in at Login. You will be asked to pick a username the first time — it is how you appear on the feed.
Choose a plan on Pricing. Creating requires a paid plan — the cheapest routes in are 3D Lite at $4.99, a $5 credit pack, or the book bundle.
Confirm your balance is showing on Profile.

CAUTION — A new account may show a credit balance before you have paid. Those credits are not spendable. Browsing is free; generating is not.

3.2 — MAKE YOUR FIRST VIDEO

Goal: one finished clip, with the least wasted credit.

Open Generator.
Choose a budget model first — Veo 3.1 Lite is the cheapest per second (47 credits for 5s). See Every model.
Set duration to the shortest option (usually 5s). Cost is per second.
Set the shape — 16:9 landscape or 9:16 vertical — before you render, not after.
Write the prompt. Say the subject, the action, and the camera move. "A red fox walks through snow, slow push in" beats "a fox".
Read the credit cost on the button. That number is exactly what you will be charged.
Press generate and leave the tab open while it works.
Happy with the composition? Re-run the same prompt on a premium model.

NOTE — The prompt is the hard part, not the model. Getting it right cheaply and then paying once is the single biggest saving available to you.

CAUTION — Closing the tab mid-render means you may not receive the result, even though the work was done.

3.3 — BUILD A CHARACTER

Goal: a locked character with a @handle you can reuse anywhere.

Open Characters. This is the only page that creates a usable character.
Give them a name. A handle is generated from it — for example @maya.
Provide a face: upload a photo, or generate one from a description.
Optionally add a body shot and an outfit image. More angles means better consistency later.
Fill in what you know — height, hair, eye colour, age. These feed the description sent to the model.
Press Lock character. You are taken to the Generator with the handle ready.

NOTE — Creator Hub displays your locked characters, and Avatar Looks handles outfits — but neither of them issues a handle. Always create on Characters.

CAUTION — Your plan limits how many characters you can keep. If you hit the limit, delete one or upgrade — see Pricing.

3.4 — PUT A CHARACTER IN A VIDEO

Goal: the same face across different scenes.

Open Generator.
Choose a model that supports reference-to-video — Seedance 2.0, Wan 2.7 or Vidu Q3.
Type the handle directly in your prompt: "@maya walks through a night market, neon reflections".
The character's reference images are attached automatically. Confirm they appear before rendering.
Check the cost, then generate.

NOTE — Consistency comes from reference images, not from describing the face again in words. Adding "brown hair, green eyes" to a prompt that already has @maya does not help and can conflict.

3.5 — TWO CHARACTERS IN ONE SHOT

Goal: two recognisable people sharing a scene.

Lock both characters first — procedure 3.3.
In Generator, select Wan 2.7 reference-to-video. It is built for this: one reference image per character.
Name both handles in the prompt and say what each is doing: "@maya hands a letter to @sam across a table".
Generate.

CAUTION — Models that take a single reference image will blend two faces into one person. If both must stay distinct, use Wan 2.7.

3.6 — MAKE AN IMAGE

Goal: a still, cheaply.

Open Image Studio.
Start on FLUX Schnell — 1 credit, the cheapest image model — to test the prompt.
Set the shape, write the prompt, check the cost, generate.
Re-run the same prompt on Flux 2 Pro or GPT Image 2 for the finished version.
Saved images appear in the library beneath the composer, and on your Profile.

NOTE — Every generation is saved private to your account. Publishing to the feed is a separate, opt-in step — see 3.14.

3.7 — EDIT AN EXISTING IMAGE

Goal: change something in a picture you already have.

In Image Studio, upload the image or pick one from your library.
Choose an edit model. GPT Image 2 is the default; Nano Banana is better at keeping a person looking like themselves; Grok Edit takes several references at once.
Describe only the change — "put her in a red coat" — not the whole scene again.
Generate. The original is untouched; the edit is saved as a new image.

NOTE — For a face swap specifically, use Face Swap. It needs two images: the face to use, and the photo to put it into.

3.8 — STORYBOARD TO FINISHED FILM

Goal: several shots that cut together as one piece.

Open .
Write your premise in a few sentences. Press Generate storyboard — this is free; it plans shots in your browser and costs nothing.
Review each shot. Edit the prompt, and the camera move, on any card. The line underneath shows exactly what will be sent to the model.
Press ✨ Improve on any shot to add framing, lens, lighting and colour terms it is missing. Also free.
Press Generate sheet to render the still frames. You are charged per frame.
Choose a video model, then press Make video. Each frame is animated and the clips are stitched in order.

NOTE — Planning and improving cost nothing. You only pay for frames and clips, so iterate on the plan as much as you like before rendering anything.

3.9 — MAKE A CLIP LONGER

Open .
Select the clip to continue.
Describe what happens next. The model continues from the final frame.
Check the cost and generate.

NOTE — Pixverse v6 is the cheapest way to extend a clip; Wan 2.5 also supports it.

3.10 — MAKE A CHARACTER SPEAK

Open Talk.
Choose the face — a locked character, or upload a photo.
Provide the words: type text to be spoken, or upload your own audio.
Pick a voice if you are using text. Voices come from ElevenLabs; you can also clone one on Sound.
Check the cost and generate.

CAUTION — Do not clone a real person's voice or likeness without their permission. See Terms.

3.11 — MAKE A 3D MODEL

Open 3D Studio.
Provide a prompt, or an image — a clean subject on a plain background works best.
Start with Hunyuan3D Rapid to check the shape is right.
Re-run on Hunyuan3D Pro or Tripo H3.1 for the version you will keep. Tripo gives the friendliest geometry for other 3D software.
Download the file.

NOTE — 3D needs the 3D add-on, from $4.99 — no full upgrade required. See Pricing.

3.12 — MUSIC AND SOUND EFFECTS

For a full track, open Music and describe genre, instruments, mood and tempo.
For a single effect or spoken audio, open Sound.
Check the cost and generate. Audio is saved to your library like any other asset.
Combine audio with video in Editor.

3.13 — UPSCALE

Open Upscale.
Upload or select the image or video.
Pick the target size, check the cost, run it.

NOTE — Upscaling a good 720p render is usually cheaper than generating in 4K, and often looks just as good.

3.14 — PUBLISH TO THE FEED

Everything you make is private by default.
To share, use the publish action on the result, or from your library.
It then appears on Feed under your username.

CAUTION — Publishing is public. Do not post anything containing a real person without their consent, or anything that breaks the Terms.

3.15 — BUY CREDITS OR CHANGE PLAN

Open Pricing.
Credit packs are one-off and never expire — good if you generate in bursts.
Plans give a monthly allowance and cost less per credit — good if you generate regularly. Annual billing saves about 17%.
Payment is handled by Stripe. Receipts and card changes are in the billing portal on Profile.

NOTE — Credits from packs stack on top of a plan. Have a partner code? Enter it on Profile.

3.16 — CANCEL OR DELETE

Open Profile.
Cancel subscription stops the next payment. You keep your plan and remaining credits until the end of the period you have already paid for.
Delete profile is permanent and removes your generations and characters. It warns you first and asks you to type DELETE.

CAUTION — If you only want to stop paying, cancel. Deleting destroys your work as well. Details on Cancellation.

3.17 — USE THE CREATOR HUB

Goal: find and reuse what you have already made.

Open Creator Hub. Characters you have locked appear here automatically.
Click any character to load it as a source for a new generation.
Items you save here are pictures you can reuse as a source image — they do not attach to a character as a wardrobe.

CAUTION — The Hub displays characters; it does not create them. A character only gets a usable @handle when it is locked on Characters — see 3.3.

3.18 — EDIT AND ASSEMBLE CLIPS

Goal: turn several clips into one finished piece.

Open Editor.
Add the clips you want, in order.
Trim each one to the part you actually want to keep — most generated clips have a weak first and last half-second.
Add audio: a track from Music, effects from Sound, or a voice line from 3.10.
Export the finished piece.

Session indicator. Above the track heads there is a small light. Red — EDIT means the timeline is live and you can edit. After two quiet minutes it turns green — STANDBY: playback, seeking, zoom and export still work, but edits are held so a stray click cannot change anything. Nothing is saved, closed or lost — your timeline is exactly as you left it. Move the mouse or press a key and it goes straight back to red.

NOTE — Editing and trimming cost no credits. Only generating does.

3.19 — BUILD A WORLD

Goal: several shots that share one consistent place.

Open Worlds.
Describe the place once — architecture, era, light, weather. This description is reused for every shot.
Add reference images if you have them; a world holds together far better with a visual anchor than with words alone.
Add shots. Each one inherits the world description, so you only write what changes.
Render. Shots can chain, so the end of one continues into the next.

NOTE — Put your characters in the world with their @handle — see 3.4.

3.20 — AGENCY WORKSPACE

Goal: keep work for different clients apart.

Open Agency.
Create a workspace per client or campaign.
Generations made in a workspace stay grouped there, so nothing bleeds between clients.
Use Projects to organise deliverables inside a workspace.

NOTE — This is aimed at the Agency plan, which carries the largest monthly credit allowance. See Pricing.

3.21 — YOUR LIBRARY

Goal: find, reuse and download anything you have made.

Open Profile. Everything you generate is saved here automatically and privately.
Filter by type — image, video, audio, 3D.
Click any item to open it, download it, or use it as the starting point for something new.
Your credit balance, plan and billing portal are on the same page.

NOTE — Saving is automatic. Publishing is not — sharing to the feed is always a separate, deliberate step (3.14).

3.22 — THE DIAMOND PARADOX BUNDLE

Goal: the novel, plus credits, for $9.99.

Sign in first. The book, the credits and the discount code all attach to your account.
Open The Diamond Paradox and buy the bundle.
The credits land on your balance and unlock creating — this is one of the cheapest routes in.
Download the book from your Profile.

CAUTION — Stripe accepts one code per order, so the bundle's discount code cannot be combined with another promotion on the same purchase.

Symptom	Likely cause	Action
Render failed	The model refused or errored.	You were not charged — credits return automatically. Reword and retry.
Balance looks wrong	A render in progress holds its credits until it finishes.	Wait for it to finish. If it still does not add up, open the help bubble — every charge is timestamped and can be checked.
Character does not look the same	Not using a reference-capable model, or describing the face in words instead.	Use Seedance 2.0, Wan 2.7 or Vidu Q3 and the @handle. See 3.4.
Two characters merged into one	The model only accepts one reference image.	Switch to Wan 2.7 — see 3.5.
Video came out the wrong shape	Aspect ratio was set after generating, not before.	Set the shape first, then render.
Prompt was refused	The provider's safety system rejected it.	Reword it. Refusals before rendering are not charged.
"Creating needs a paid plan"	Your balance is granted credits, not purchased.	Any purchase unlocks generating — see Pricing.
Taking much longer than usual	Premium models are slower; occasionally a job is parked for a manual fix.	Leave the tab open. You will get the result or your credits back.
Cannot find a character you made	It was made somewhere other than the Characters page.	Only Characters issues a usable handle. Re-lock it there.

DIRECTING SEEDANCE 2.5 — CAMERA & PROMPT CONTROLS

Seedance 2.5 follows film language written in plain English. These controls come from the model maker's own documentation and are proven on JOVON.

Camera moves it understands

Shot scale: extreme close-up · close-up · medium close-up · medium shot · wide shot · establishing shot.
Angles: eye level · low angle (power) · high angle (vulnerability) · top-down · first-person.
Moves: slow push-in (rising emotion) · slow pull-out (release, endings) · smooth lateral move · tracking shot · orbit · handheld.
One primary camera move per shot. A tight close-up plus a big rotation is unstable — go wider or plan a cut.

Fancy terms: name it, then describe it

For niche moves (dolly zoom, bullet time, rack focus, speed ramp, whip-pan), write the term and what it looks like: “rack focus: shift focus from the foreground leaves to the person behind them — the leaves go soft while the face sharpens.” A model that knows the term takes the shortcut; one that doesn't follows your description.

The prompt blocks that matter

Reference binding: say what each input controls — “Use the woman in @Image1 as the main character.” In Reference mode, cite inputs in upload order: @Image1, @Image2, @Video1, @Audio1.
One action per shot: one observable event, not three.
End state: finish stages on something visible — “End state: she holds the cup in her left hand; the door is closed.” “She feels relieved” is not an end state; “her shoulders drop and the frown clears” is.
Constraints: protect what matters — “Keep the label readable; no subtitles.”
Multi-shot films: order events as Shot 1 / Shot 2 / Shot 3 in one prompt.

Reference mode superpowers

Up to 30 reference images (characters, products, scenes, styles), 10 reference videos, and 10 audio clips — a music track can drive the visual pacing and lip-sync. For a consistent person, use a clean face close-up plus a separate full-body image and name their role. Editing an existing clip? Phrase the prompt as an edit of @Video1 (“In @Video1, replace the background with… keep the people and camera unchanged”) — duration follows the source clip.

What gets a render blocked (learned the real way)

• Explicit gore (“blood drips…”) — write “battle-worn” instead.
• Real directors', artists' or brand names (“Ridley Scott-esque”, “Superman”) — describe the look instead: “mythic, high-contrast epic”, “blue superhero suit with a red cape”.
• Uploading photos or videos of real people's faces — use JOVON-generated characters instead.
A blocked render is refunded automatically.

FRAMES vs REFERENCES — WHAT EACH IMAGE SLOT IS

Every image you upload does one of two completely different jobs. Getting these mixed up is the single most common reason a render comes back wrong.

1. A frame is a moment

The picture is a specific instant of the finished video. There are only ever two, and the order means time.

Mode	What you give it
Image → Video	1 image = the first frame. The video starts on exactly that picture.
Start + End	2 images = the first frame and the last frame. The model invents the motion between them.
Video → Video	1 video = the clip being edited. Anything you do not mention stays as it is.

2. A reference is a subject

In Reference mode the pictures map to nothing in the timeline. They tell the model who and what should appear — it invents every frame itself. This is why one model can take 30 of them: they are not 30 moments, they are 30 things to stay consistent about.

Order still matters, but for a different reason: it sets the @Image number your prompt points at, and the model leans hardest on @Image1. Remove one and the rest renumber — check your prompt still points where you meant.

A worked example — 7 images, one video

On Kling o3 Pro Ref:

@Image1 your model, front ← the anchor
@Image2 same model, side
@Image3 same model, back
@Image4 the jacket, flat lay
@Image5 the jacket, on a hanger
@Image6 the location, wide
@Image7 the location, close

Prompt: “@Image1 walks through @Image6 wearing @Image4, camera pulls back.” Seven images, one video, and not one of them is a frame.

What each model's references are for

Model	Takes	What they are for
Seedance 2.5	30 images 10 videos 10 audio	Character, style and scene references, video editing, or all combined. Audio-only referencing works here and nowhere else — hand it a music track or a voice and it generates to match.
Seedance 2.0 / Mini / Fast	9 images 3 videos 3 audio	Same idea, smaller limits. Mini is the cheapest way to mix image + video references.
Wan 3.0 Prime	10 images 5 videos 5 audio	“All-in-One Reference” — pixel-level identity, voice and space alignment. The cheapest way to hold a subject consistent.
Kling o3 Pro / Std Ref	7 images	Character, prop or scene references. With a source video the image limit drops to 4.
Wan 2.7	4 images	Character-driven video with multi-subject support.
Vidu Q3 / Q3-Mix	4 images	Consistent subjects across intelligent camera switching between positions.

Which one do I want?

• You have the exact picture the video should open on → Image → Video.
• You have the opening and closing picture → Start + End.
• You want the same person, product or place to appear reliably, and you do not care which exact frame → Reference.
• You have a clip and want one thing changed → Video → Video.

TIPS & CONTROLS, MODEL BY MODEL

The real controls each model accepts on JOVON — durations, shapes and the one tip that gets the most out of it.

Seedance 2.5 — 4–15s (up to 30s in Reference mode) · 480p/720p · sound generated automatically. Tip: it follows film language best of all models — use the Directing guide above. In Reference mode a music clip can drive the pacing.

Seedance 2.0 — 4–15s · 480p up to real 4K · up to 9 reference images. Tip: the reliable workhorse; the only lane where JOVON 4K is receipt-proven. Start+End frames work here too.

Seedance Mini — 5/10/15s. Tip: cheapest Seedance — use it to draft a shot before spending on the big models.

Kling v3 Turbo — 5s clips. Tip: strongest at fast motion and action beats; keep prompts to one action.

Kling o3 (Video Edit / VFX) — edits an existing clip up to 10s. Tip: describe the CHANGE and what must stay: “replace the background with X, keep the person, their motion and the camera unchanged.” Original sound is kept.

Veo 3.1 Lite — 4/6/8s ONLY · 16:9 or 9:16 ONLY. Tip: the best price on the menu for quick drafts — but pick a supported length and shape or nothing renders.

Pixverse v6 — 1–15s · 360p/540p/720p/1080p quality tiers · Start+End frames. Tip: the flexible-length value pick; great for short loops and transitions between two exact frames.

Vidu Q3 — image, reference and Start+End modes · keeps up to 7 subjects consistent. Tip: the multi-character alternative when a scene has a crowd.

Wan 2.7 — the sturdy all-rounder · reference and video-edit modes. Tip: a safe default when you're not sure which model fits.

Wan 2.5 Fast / Wan 2.6 — 5/10s and 5/10/15s. Tip: Fast for speed, 2.6 for a quality bump at mid price.

Grok Imagine — 5s stylized shots. Tip: leans arty — give it bold visual styles rather than realism.

Lip Sync — a video + a voice clip (30s max). Tip: Standard (veed) is 6× cheaper than Pro; use Pro only when the face is large in frame. Wider framing syncs more reliably than extreme close-ups.

Two rules that improve every model

Write results you can see, not moods. Not “make it tense” — instead: “her jaw tightens, she stops mid-step, her eyes flick to the door.” An instruction you can't check on the output can't be fixed either.
Keep global rules global. Style, world and character locks go at the TOP of the prompt; per-moment events go in their shot. A style note buried in shot 3 stops applying by shot 4.

USE JOVON FROM YOUR AI (MCP)

Connect your own Claude to your JOVON account and generate by chatting: “Claude, make me a 5-second video of a lighthouse in a storm on JOVON.” Claude starts the render, waits for it, and hands you the finished video — billed to your credits, saved in your Library.

Step 1 — get your URL. Profile → Connect Claude → Get my connection URL. It looks like https://jovon.ai/mcp/jvk_xxxxxxxx. Your key is inside the URL, so no API keys, headers or environment variables are needed anywhere — if a tool accepts a remote MCP URL, it works.

Step 2 — paste it into whichever AI you use.

Claude (claude.ai, Desktop): Settings → Connectors → Add custom connector → paste the URL. Requires a paid Claude plan.

Claude Code — one line in your terminal:

claude mcp add --transport http jovon <your URL>

ChatGPT: Settings → Connectors → add a custom MCP connector and paste the URL. Availability depends on your ChatGPT plan — if you do not see Connectors, your plan does not expose them yet.

Cursor, Windsurf, VS Code, Zed, Goose, Gemini CLI, Codex CLI and most other MCP clients read a small JSON config. Add JOVON as a remote HTTP server:

{
  "mcpServers": {
    "jovon": {
      "type": "http",
      "url": "https://jovon.ai/mcp/jvk_your_key_here"
    }
  }
}

Where that file lives differs per app — Cursor uses ~/.cursor/mcp.json, VS Code uses .vscode/mcp.json, and several CLIs offer an mcp add command that writes it for you. Some older clients use "transport" instead of "type". Check your app's MCP docs for the path; the URL is the only part that comes from JOVON.

What your AI can then do: jovon_generate_video, jovon_generate_image, jovon_check_render and jovon_credits. Renders bill your credits, land in your Library, and pass the same content filters and refund rules as the buttons on the site — there is no separate quota and no second bill.

Your connection URL is your password — anyone holding it can spend your credits. Never paste it into a shared chat, a screenshot or a public repo, and regenerate it on your profile the moment you suspect it leaked. Full walkthrough: jovon.ai/claude

EVERY MODEL, AND WHAT IT'S BEST FOR

Every model below is live in the studio right now — and every new account gets 25 free images + 3 free videos (5-second starter films) to try the studio. You never have to guess the price — the exact credit cost is shown on the button before you click, and it is the same number the server charges.

◆ Budget · cheapest per second ◆ Everyday ◆ High ◆ Premium · the dearest, use when it matters

VIDEO

Model	Band	Best for
Veo 3.1 Lite	Everyday	Strong prompt-following for the money. Also does start-frame → end-frame, so you can control where a shot begins and ends.
Pixverse v6	Everyday	Stylised and animated looks. Carries audio, and can extend a clip.
Wan 2.5 Fast	Everyday	Fast turnaround at 5 or 10 seconds. Good for social-length clips.
Kling v3.0 Turbo	High	The v3 look without the v3 price. A good step up when a draft is nearly right.
Vidu Q3	High	Clean movement. Q3 is the reference-to-video one — feed it up to four images to hold a subject steady.
Wan 2.6	High	5, 10 or 15 seconds in a single shot, with multi-shot support.
Seedance 2.0 Mini	High	Most of Seedance 2.0's quality for noticeably less.
Seedance 2.5 NEW	Premium	The new flagship. Single shots up to 30 seconds with built-in sound — voices, effects and music generated with the picture. The longest, most complete clips on JOVON.
Seedance 2.0	Premium	Excellent natural motion and prompt adherence. Reference-to-video takes up to 9 images — the best choice for keeping a character consistent.
Wan 2.7	Premium	The two-character model. Its reference mode takes one image per character, so two people can share a scene and both stay recognisable.
Veo 3.1	Premium	The photoreal flagship — very strong at following a complicated prompt exactly, with lifelike physics and light.
Video Edit (Kling o3) NEW	High	Video → Video. Feed it your own clip (up to 10 seconds) and describe the change — it repaints the look while keeping your motion and timing. Standard and Pro tiers; billed on your clip’s length.
Grok Imagine Video	High	Text or image to video, and can extend an existing clip.
Vidu Q3-Mix	Premium	Start-frame → end-frame transitions and mixed references — blend several source images into one moving shot.

IMAGE

FLUX Schnell — 1 credit, near-instant. The cheapest image on the site: test prompts and make volume here. Free starter images use this model.
FLUX 2 Pro — sharper detail and better text rendering, for finished work.
Imagen 4 / Imagen 4 Ultra — Google’s image line: excellent realism and lighting. Ultra for the shots that matter.
MAI Image 2.5 — a strong all-rounder at a mid price.
GPT Image 2 — best all-round prompt understanding, and the standard editor.
Nano Banana 2 / Pro — the character-consistency specialists: strongest at editing an existing image without losing the subject.
Seedream v4 / v5 Lite — image-editing lanes for restyling what you already have.
Face Swap — one face into another photo. Needs two images: the face, and the photo to put it in.

3D

Hunyuan3D Rapid — fast, cheap. The one to try an idea with.
Hunyuan3D Pro — cleaner geometry, better detail. For something you'll actually use.
Seed3D v2 — image → 3D only (that is why it greys out in Text → 3D): give it a photo, get a model.
Tripo H3.1 — good topology; the friendliest output if you're taking it into other 3D software.

3D needs the 3D add-on — from $4.99, no full upgrade required. Models download as standard files you can open elsewhere.

SOUND & SPEECH

ElevenLabs — speech from text, and voice cloning. The most natural voices available here.
Sound FX — individual effects from a description, on Sound.
Music — full original tracks, on Music.
Lip sync (Veed / Sync v3) — make a face match audio. Veed is the budget option; Sync v3 is far dearer and more accurate.

How to choose without wasting credits

Draft on a budget model first. Get the prompt right on FLUX Schnell (images) or Seedance Mini (video), then re-run the same prompt on a premium model. The prompt is the hard part, not the model.
Only pay for length you need. Cost is per second, so a 10-second clip costs roughly twice a 5-second one.
Reference-to-video is what keeps a face consistent — not a better prompt. Seedance 2.0 (up to 9 images), Wan 2.7 (one per character), Vidu Q3 (up to 4).
4K is rarely worth it unless something is going on a big screen. Upscaling a good 720p render is usually cheaper.
A failed render costs you nothing. Credits are held while it runs and returned automatically if it fails.

HOW TO USE JOVON

START HERE — THE BASICS

EVERY PAGE, EXPLAINED